Skip to content

Overview

Here you find general descriptions of how to run jobs on LUMI, i.e. how to run your scientific software using the job scheduler on LUMI. In this section, you also find general information about the LUMI environment, as well as the LUMI web interface that you can use to run your applications on LUMI instead of traditionally using a terminal. If you are looking for ways to install your software on LUMI or advice for a specific application, consult the software section instead.

When you log in to LUMI, you get access to one of the login nodes. These login nodes are shared by all users and are only intended for simple management tasks, e.g.

  • compiling software (but consider allocating a compute node for large build jobs)
  • submitting and managing scheduled jobs
  • moving data
  • light pre- and postprocessing (a few cores / a few GB of memory)

All compute heavy tasks must be submitted through the job scheduler such that they are run on the compute nodes in LUMI-G/LUMI-C/LUMI-D.

The job scheduler used on LUMI is Slurm. To get started using Slurm on LUMI, read the Slurm quickstart guide.

To manage the load on the login nodes and to make sure that any hanging processes get cleaned up, the walltime of processes on the login nodes is limited to 336 hours (2 weeks). After a process reaches this time limit, it is first softly killed, before a hard kill is sent, to ensure that if a process still has to write some data, it can do so properly (like shell histories). Also to ensure that the policy for not running heavy tasks on login nodes is followed, the maximum CPU time of a job on a login node is set to 24 CPU-core-hours. This is long enough for software builds or reasonable data transfer to LUMI-O not to be interrupted, yet short enough that a compute job using the 16 cores that it can use at 100% would get killed after 1.5 hours.

Warning

All tasks not adhering to the above fair use rules for the login nodes will be terminated without warning. The walltime of processes is limited to 336 hours (2 weeks), and the maximum CPU time of jobs is 24 CPU-core-hours.

Also, you may want to familiarize yourself with the LUMI environment:

  • Read the module environment page to learn more about how to use the module system on LUMI to find already installed software and to manage your own software installations.
  • Read the software stacks page to learn more about the software is already centrally installed on LUMI.
  • Read the data storage options page to learn more about where to store your data.
  • Read the billing policy page to learn more about how you are billed for your use of LUMI.